Share your thoughts, 1 month free Claude Pro on usSee more

Vision-Language Compositionality on VisMin Hard (test)

84.47Text Score (T)

Qwen2.5-VL-7B + FOCUS

Updated 1mo ago

Evaluation Results

Method	Links
Qwen2.5-VL-7B + FOCUS 2025.08		84.47	61.44	56.99
Qwen2.5-VL-7B 2025.08		84.37	53.26	49.97
Qwen3-VL-4B + FOCUS 2025.08		84.11	58.31	54.32
Qwen3-VL-4B 2025.08		83.62	49.62	46.78
InternVL3-8B 2025.08		82.86	47.66	44.8
Qwen3-VL-8B 2025.08		82.86	56.65	52.95
Qwen3-VL-8B + FOCUS 2025.08		82.77	59.13	55.01
InternVL3-8B + FOCUS 2025.08		82.54	52.66	48.89
LLaVA-OneVision-7B 2025.08		80.61	44.42	38.4
LLaVA-OneVision-7B + FOCUS 2025.08		80.44	57.89	49.4
Qwen2.5-VL-3B + FOCUS 2025.08		78.2	28.86	25.29
Qwen2.5-VL-3B 2025.08		77.38	27.09	22.93
InternVL3-2B 2025.08		65.53	6.84	6.17
InternVL3-2B + FOCUS 2025.08		64.87	26.77	23.96
LLaVA-OneVision-0.5B + FOCUS 2025.08		2.2	18.41	0.59
LLaVA-OneVision-0.5B 2025.08		1.7	20.12	0.51